الملخص
Looped Transformers reuse a shared block across recurrent depths, making autoregressive decoding expensive because every generated token requires many sequential recurrent passes. Self-speculative decoders reduce this cost by drafting at an early depth and verifying at full depth, but typically bind draft computation to prefix representations from the same recurrent depth. We find that queries and keys approach their final-depth representations earlier than values, and controlled prefix-channel interventions show that mature values substantially improve shallow draft predictions. Motivated by this asymmetry, we introduce Depth-Asynchronous Self-Speculation (DAS), which decouples the depth of draft computation from the depth of verified-prefix representations it reads. Its Mature-V primitive lets shallow queries retrieve full-depth prefix values without additional recurrent computation. We further develop DAS-Wave, which combines depth-asynchronous prefix reads with carried parallel refinement, progressive block growth, and an independent full-depth verifier. Across four recurrent-model checkpoints and mathematics and code workloads, DAS-Wave achieves 4.00--6.96$\times$ mean throughput speedup over paired full-depth autoregressive decoding in the same inference stack. These results identify prefix-information depth as an effective design axis for recurrent self-speculation.
الكلمات المفتاحية
الموضوع
بيانات النشر
- المجلة
- غير متاح
- وصول مفتوح
- وصول مفتوح أخضر
اقتبس هذه المقالة
APA 7
Li, G., Su, Z., Yu, H., Jiang, J., Ren, T., Li, Z., Lu, F., Tang, M., & Yuan, C. (2026). Shallow Queries, Mature Values: Depth-Asynchronous Self-Speculation for Looped Transformers. https://omanscience.com/ar/articles/shallow-queries-mature-values-depth-asynchronous-self-speculation-for-looped-transformers
MLA 9
Li, Guanghao, et al. "Shallow Queries, Mature Values: Depth-Asynchronous Self-Speculation for Looped Transformers." https://omanscience.com/ar/articles/shallow-queries-mature-values-depth-asynchronous-self-speculation-for-looped-transformers.
شيكاغو (المؤلف–التاريخ)
Li, Guanghao, Zihan Su, Hao Yu, Jinyang Jiang, Tao Ren, Zehao Li, Feng Lu, Ming Tang, and Chun Yuan. 2026. "Shallow Queries, Mature Values: Depth-Asynchronous Self-Speculation for Looped Transformers." https://omanscience.com/ar/articles/shallow-queries-mature-values-depth-asynchronous-self-speculation-for-looped-transformers.
هارفارد
Li, G., Su, Z., Yu, H., Jiang, J., Ren, T., Li, Z., Lu, F., Tang, M. and Yuan, C. (2026) 'Shallow Queries, Mature Values: Depth-Asynchronous Self-Speculation for Looped Transformers', Available at: https://omanscience.com/ar/articles/shallow-queries-mature-values-depth-asynchronous-self-speculation-for-looped-transformers.
فانكوفر
Li G, Su Z, Yu H, Jiang J, Ren T, Li Z, et al. Shallow Queries, Mature Values: Depth-Asynchronous Self-Speculation for Looped Transformers. https://omanscience.com/ar/articles/shallow-queries-mature-values-depth-asynchronous-self-speculation-for-looped-transformers
IEEE
G. Li, Z. Su, H. Yu, J. Jiang, T. Ren, Z. Li, F. Lu, M. Tang, and C. Yuan, "Shallow Queries, Mature Values: Depth-Asynchronous Self-Speculation for Looped Transformers," https://omanscience.com/ar/articles/shallow-queries-mature-values-depth-asynchronous-self-speculation-for-looped-transformers.